Issues

• lil-bench: KV budget = num_gpu_blocks × block_size under-counts hybrid-KV vLLM; long-context cells skipped

  local-inference-lab/llm-inference-bench#21 · Filter by author Bjorgz opened on Sep 27, 2026

• Could you add a LICENSE file?

  local-inference-lab/llm-inference-bench#17 · Filter by author marcosvasili-coder opened on Sep 22, 2026

• Warmup readiness can be overwritten by timeout in the same polling iteration

  local-inference-lab/llm-inference-bench#15 · Filter by author shussekaido opened on Sep 7, 2026

• P2P detector requires registry selectors that break CUDA peer copies on RTX PRO 6000 Blackwell

  local-inference-lab/llm-inference-bench#7 · Filter by author jackzampolin opened on Jul 24, 2026

• Dataset setup for Reasoning and Speculative Decoding

  local-inference-lab/llm-inference-bench#2 · Filter by author MLKoz2 opened on Apr 2, 2026